Molecular Ecology (2007) 16, 1993– 2004
doi: 10.1111/j.1365-294X.2007.03268.x
Detecting genetic structure in migrating bowhead whales off
the coast of Barrow, Alaska
Blackwell Publishing Ltd
P . E . J O R D E ,*† T . S C H W E D E R ,*‡ J . W . B I C K H A M ,§¶ G . H . G I V E N S ,** R . S U Y D A M ,†† D . H U N T E R § and
N. C. STENSETH*
*Centre for Ecological and Evolutionary Synthesis (CEES), Department of Biology, University of Oslo, PO Box 1066 Blindern, N-0316
Oslo, Norway, †Institute of Marine Research, Flødevigen, N-4817 His, Norway, ‡Department of Economics, University of Oslo, PO
Box 1095 Blindern, N-0317 Oslo, Norway, §Department of Wildlife and Fisheries Sciences, Texas A&M University, College Station,
TX 77843-2258, USA, **Department of Statistics, 1877 Campus Delivery, Colorado State University, Fort Collins, CO 80523, USA,
††North Slope Borough, Department of Wildlife Management, PO BOX 69 Barrow, AK 99723, USA
Abstract
We develop a general framework for analysing and testing genetic structure within a migratory
assemblage that is based on measures of genetic differences between individuals. We demonstrate this method using microsatellite DNA data from the Bering-Chukchi-Beaufort
stock of bowhead whales (Balaena mysticetus), sampled via Inuit hunting during the spring
and autumn migration off Barrow, Alaska. This study includes a number of covariates such
as whale ages and the time separation between captures. Applying the method to a sample
of 117 bowhead whales, we use permutation methods to test for temporal trends in genetic
differences that can be ascribed to age-related effects or to timing of catches during the
seasons. The results reveal a pattern with elevated genetic differences among whales caught
about a week apart, and are statistically significant for the autumn migration. In contrast,
we find no effects of time of birth or age-difference on genetic differences. We discuss possible explanations for the results, including population substructuring, demographic
consequences of historical overexploitation, and social structuring during migration.
Keywords: Balaena mysticetus, bowhead whale, genetic structure, microsatellites, statistical modelling,
temporal pattern
Received 5 August 2006; revision accepted 11 December 2006
Introduction
Most previously developed methods for the assessment of
genetic variability and the concomitant identification of
mixed-stock assemblages (e.g. Smouse et al. 1990; Paetkau
et al. 1995; Rannala & Mountain 1997) require genetic data
from geographically separated breeding populations as
reference material. However, in some cases data from breeding
sites are not available because they are remotely located,
unknown, or inaccessible. For example, genetics of freshwater
eels (Anguilla) has been studied from their freshwater
habitats in Europe and North America (Avise et al. 1986;
Correspondence: Per Erik Jorde, Institute of Marine Research,
Nye Flødevigveien 20, N-4817 His, Norway. Fax: +47 37 05 90 01;
E-mail: p.e.jorde@bio.uio.no
Present address: ¶Center for the Environment and Department of
Forestry and Natural Resources, Purdue University, West Lafayette,
IN 47907, USA.
© 2007 The Authors
Journal compilation © 2007 Blackwell Publishing Ltd
Daemen et al. 2001; Wirth & Bernatchez 2001). But these
species spawn at great depths in the Sargasso Sea from
where samples are not available. Furthermore, genetic
analyses of population structure in migratory marine animals
typically require carefully planned temporal and spatial
collections of individuals to be representative (e.g. Nesbø
et al. 2000). For many organisms, including the cetaceans
considered herein, such data are not available. Instead,
samples are generally collected from animals killed near a
few land-based villages during subsistence hunts or opportunistically during commercial hunts. For such organisms,
few analytical tools are currently available, and those that
are available (e.g. structure and baps: Pritchard et al. 2000;
Corander et al. 2003, respectively) typically do not utilize
auxiliary information and may lack in statistical power in
specific applications. The methods and analyses presented
in this study directly address such problems.
The bowhead (Balaena mysticetus) is a large baleen whale
inhabiting the Arctic oceans. Five stocks are presently
1994 P . E . J O R D E E T A L .
recognized within the species’ distribution range. These
stocks were all subject to extensive commercial hunting in
the 19th century, and the Bering–Chukchi–Beaufort Seas (BCB)
stock is the only one that is certain to be recovering successfully after commercial whaling ended in 1914 (Bockstoce
1986; Bockstoce & Burns 1993; Zeh et al. 1993; George et al.
2004). Whales of the BCB stock winter in the Bering Sea and
migrate northward to summering areas, predominantly in
the eastern Beaufort Sea (Moore & Reeves 1993). Based on
Russian observations and oceanographic data, there might
be two distinct patterns in feeding migration for whales
wintering in the Bering Sea. Melnikov et al. (2004) and
Bogoslovskaya (2003) describe a delay in spring migration of
bowheads out of the Gulf of Anadyr and north through the
Bering Strait as well as sightings on the Chukotka coast in
summer and early autumn. Some animals may reside in the
Bering or Chukchi seas during summer but the numbers
and whereabouts of these are poorly known. Mating likely
occurs from March to May, possibly in wintering areas or
during the spring migration (Koski et al. 1993).
In migration, whales are hunted for subsistence by
aboriginal peoples, primarily in Alaska. Of the 10 villages that
hunt in Alaska, Barrow is the only community that harvests
bowheads both during the spring and autumn migration,
and lands about 70% of the total catch. Examination of landed
whales has provided a great deal of information about the
biology of bowheads including some data on their unique
characteristics: they are very long-lived, perhaps up to
200 years or more, and attain sexual maturity at a rather
late age of approximately 20 years (George et al. 1999).
The International Whaling Commission (IWC) has managed BCB bowheads as a single stock, and has developed
a management plan for Alaskan aboriginal subsistence
harvest based on the single-stock hypothesis. A general
management concern is the risk of depleting genetically
distinct population components when exploiting the stock
under the assumption of a single population. Different
populations may coexist temporally, but if they do so
during the subsistence hunt or commercial harvest, as is
typically the case, then stock management needs to take
this into account (Hoelzel 1998). Until recently, nearly all
available evidence suggested that a single-stock hypothesis
was the most reasonable one (Rugh et al. 2003). However,
the availability of new genetics data has motivated further
study of potential substructuring of BCB bowheads, uncovering some reasons to question the single-stock hypothesis.
Aside from its management implications, the issue of potential
structure also has important implications for understanding the biology and demography of this species.
Multilocus microsatellite DNA polymorphisms have
seen widespread use as a powerful tool for detecting population structure in a wide range of species, including whales
and dolphins (Hoelzel 1994; references therein). Previous
genetic studies of the BCB stock of bowhead whales have
been restricted largely to testing for effects of putative population bottlenecks caused by commercial whaling (Rooney
et al. 1999a, 2001). These studies have uncovered high levels
of genetic variability in BCB bowheads, but have yielded
limited information bearing on the issue of potential substructuring of the stock.
In this study, we analyse microsatellite DNA polymorphisms in migrating bowhead whales off Barrow, Alaska,
with the objective of assessing to what extent there is a
genetic substructure within the stock. As an aid in the analysis,
we develop a statistical framework that is based on genetic
differences within pairs of whales from the migratory
assemblage (cf. Rousset 2000). The method is general and
may be applicable to other migratory species also, even
when no geographical reference samples are available.
Materials and methods
The data for this study are from Hunter (2005) which were
reported to the International Whaling Commission by
Bickham et al. (2004). All individuals were genotyped anew
relative to an earlier screening that contained a subset of the
present sample (Rooney et al. 1999a; Rugh et al. 2003). The
rescreening was done in order to ensure consistent allele
calling among older and new samples. The data refer to
samples of bowhead whales (n = 207) obtained from 1983
through 2003 and scored at up to 12 microsatellite loci: Ev1
and Ev104 (Valsecchi & Amos 1996); Gata28 (Palsbøll et al.
1997); Tv7 (Rooney et al. 1999b), Tv11, Tv13, Tv14, Tv16,
Tv17, Tv18, Tv19, and Tv20 (Rooney et al. 1999a). For the
present analysis, we excluded two loci (Tv7 and Tv18) that
were found to be problematic during preliminary analyses,
displaying large deficiencies of heterozygotes and indicating
segregation of null alleles (Tv7) and short allele dominance
(Tv18), respectively. Tissue quality varied among samples
and not all individuals could be scored at all loci. We chose
to exclude all individuals that failed scoring at more than
one microsatellite after repeated attempts. The resulting
data set used in the present analyses included 10 microsatellite
loci scored in 134 bowhead whales, most of which were
caught at Barrow (n = 117), or from other Alaskan communities
(Point Hope, n = 3; Wainwright, n = 1; Nuiqsut, n = 1;
Kaktovik, n = 1), from St Lawrence Island (Gambell, n = 4;
Savoonga, n = 5), or from Russia (Chukotka, n = 2) (Fig. 1).
Individuals were caught during the spring (April–June)
and autumn (September–November) migration in approximately equal proportions (Fig. 2). The annual sample sizes
are not balanced (cf. Table 4), and we discuss potential
consequences for our analysis below.
Statistical model for genetic structure
The model that we propose is based on pairwise genetic
differences among individuals, following Rousset (2000).
© 2007 The Authors
Journal compilation © 2007 Blackwell Publishing Ltd
B O W H E A D W H A L E G E N E T I C S T R U C T U R E 1995
Fig. 1 Map of the seasonal distribution (coloured areas) and migration routes (arrows) of BCB bowhead whales. Yellow dots denote
whaling villages. Whales pass Barrow and are hunted there during both the spring and autumn migrations.
Fig. 2 Date of landing of individual whales
in different years. Open circles: whales
landed at Barrow; +: whales landed at other
villages. Points are jittered for clarity.
© 2007 The Authors
Journal compilation © 2007 Blackwell Publishing Ltd
1996 P . E . J O R D E E T A L .
The method takes advantage of the timing of individual
catches as whales pass Barrow, the primary sample site,
during their northward (spring) and southward (autumn)
migrations (Fig. 1). We analyse data from spring and
autumn separately to allow for possible differences in
migratory behaviour and structure in the two seasons.
Calendar time of migration varies from year to year, and
we consider only pairs of whales taken in the same year.
Hence, data from spring and autumn are analysed
separately, and there is stratification by sampling year. No
temporal pattern should emerge if all individuals belong
to the same biological population, assuming that the
population is genetically well mixed. However, the
population may not be well mixed for several reasons. For
example, the sampling interval (20 years) is substantial
and notable genetic change (e.g. random drift) may have
occurred during the interval if the effective bowhead
population size is small. Also, remnants of earlier (historical)
drift may still exist in the population in the form of genetic
differences among age classes (cf. Jorde & Ryman 1995;
Ryman 1997), especially considering the exceptional longevity
of this species. We used available length data to estimate
age and test whether there is any effect of time of birth on
the genetic differences among individual bowheads.
Patterns in pairwise genetic differences are detected
using a statistical model with genetic differences between
whales in pairs as the dependent variable and various
covariates related to timing of sampling and birth as independent variables (equation 2). Genetic difference in a pair
of whales was estimated from the microsatellite genotypes
following the logic of Rousset (2000). Briefly, for each locus
we estimated the probability that two individuals, i and j,
carry the same allele (‘same’ in the meaning ‘alike in state’,
not necessarily identical by descent). This was done by comparing the two gene copies carried within an individual with
the two copies carried within the other individual in the
pair and recording if they are the same (identity, I = 1) or
different (I = 0). For each locus, l, and pair of individuals, ij,
there are four such comparisons, with each of the two
genes in individual i being compared to each of the two
genes in individual j. We calculated an average similarity
score Iijl, over the four comparisons in this pair and locus.
This score was contrasted to the frequency of identity of
genes within individuals, yielding an estimate of pairwise
genetic difference between two individuals (cf. Rousset
2000):
aij =
1
Lij
∑
l
hl − I ijl
1 − hl
(1)
Here, hl is the average proportion of homozygous individuals
at the lth locus (averaged over all sampled individuals
from this season and not just the two individuals under
consideration), and Lij is the number of loci that were
simultaneously scored in both individuals i and j. The
calculations were repeated for all pairs of individuals. We
note that in the absence of any population substructuring,
genetic similarity within individuals (hl) should on average
equal that among individuals (Iijl) and the expected value
of aij is zero. It remains zero when further conditioning on
covariates not causing genetic segregation. With departure
from Hardy–Weinberg equilibrium (e.g. inbreeding), the
expected value might be different from zero, but should still
not be affected by nonsegregating covariates.
Age of whales was estimated from their measured length
by the von Bertalanffy growth function, using growth parameters from Rosa et al. (2004), separately for each sex. From ages
thus estimated and from known years of sampling, we calculated the number of years yij between birth and the mean
year of birth mij for each pair of individuals. With dij denoting
the number of days between sampling in the pair, and focusing only on pairs caught in the same year and season and
at the same locality (Barrow), we fitted the additive model:
aij ~ s(dij ) + s (yij ) + s(mij ).
(2)
Model (2) is denoted d + y + m for short, while d + y, d + m,
and y + m denote submodels leaving out one of the
covariates m, y, and d, respectively. In equation 2, the
genetic difference between individuals (aij) is the response,
and s() are smoothing spline functions with degrees of
freedom chosen by cross-validation.
The models were fitted using the S-plus function GAM
(Venables & Ripley 1999) with weights proportional to the
number of scored loci (Lij) in each pairwise comparison.
We did not separate between the sexes in the model as preliminary analyses indicated that there is no effect of sex on
a. Indeed, there are no detectable allele frequency differences
between the sexes, and the average FST (Weir & Cockerham
1984) between males and females is negative (−0.003). Sex
was therefore dropped from further consideration.
The statistical significance of genetic patterns is obtained
by analysis of variance, with null distributions calculated
from randomly permuted data. When testing for age effects
( y and m) in models only including such effects, the variance
of fitted values for observed pairs is compared to the distribution of variance of fitted values in replicated simulated
data where the response is as observed and the birth years
are randomly permuted within season and year. The same
approach is used to test if days apart (d) has a significant
effect on genetic difference in a model without age effects,
by randomly permuting day of catch within season and
year. To test whether there is a partial temporal effect on
genetic difference, days of catch are randomly permuted
while year of birth is kept as observed. The partial effect of
days apart was tested also in the d + y model. The distribution of variance of fits to simulated data is a null distribution for the observed fits variance.
© 2007 The Authors
Journal compilation © 2007 Blackwell Publishing Ltd
B O W H E A D W H A L E G E N E T I C S T R U C T U R E 1997
The effects of factors d, y and m on a are analysed as
follows. First, we fit model (2) and all its six submodels
(i.e. d, y, m, d + y, d + m, and y + m) for each season separately.
The fit of each model is measured by the P value of R2 =
var(â)/var(a), where â is the fitted value for the model to be
tested, and where the variance is taken over the paired
individuals (within year). The null distribution of R2 is estimated from 5000 replicate simulations. In each replicate,
the vectors of day of catch and year of birth are randomly
permuted over individuals within year. The model is then
fitted and a simulated null value of R2 is calculated from
the variance of predicted genetic differences, var(â), in the
simulation. We concentrate on models that significantly
explain some of the variation, as expressed by R2, and which
include the term y since some genetic drift is expected in
any finite population.
To test for partial effect of days apart, d, when controlling for age-related (y and m) effects, a new set of 5000
simulations was performed. In this test, only day of catch
is randomly permuted within year before the model is
fitted and R2 is calculated. In addition to the resulting
P value of R2, a simultaneous P value based on the confidence bands discussed below is also calculated. The
analysis of variance test is more powerful since the
confidence band method really consists of combining
separate tests for each value of d through a Bonferroni
type construction.
The shape of the partial effect curve s(dij) in model (2),
with or without the covariate m, might give a hint of a
potential genetic segregation in the population. A graph of
the estimated partial effect curve of days apart is obtained
as the predicted genetic difference within pairs taken 0, 1,
2, … days apart, assuming difference between birth years
and mean birth years equal in all the hypothetical pairs,
and equal to their medians in the sample. The following
variation of the method of Beran (1988) is used to obtain
simultaneous null-bands for the partial-effect curves. A
95% point-wise null-band is first calculated from simulated partial effect curves, using 5000 replicate permutations, with day of catch randomly permuted within season
and year and with genetics and year of births as observed.
At d days apart, the point-wise 95% interval is (qd – lowd, qd
+ highd) where lowd, qd and highd are, respectively, the 2.5%,
50% and the 97.5% percentiles of the simulated curves at d
days apart. Simultaneous null-bands of confidence are
obtained by replacing lowd by x*lowd and highd by x*highd
where x ≥ 1 is the multiplier required to contain the entire
simulated curves in 95% of the simulations. Similar to the
construction of these confidence bands we also adjusted
the factor x to obtain a null-band that just contained the
observed partial effect curve at all points. The fraction of
the null-simulated effect curves that were not entirely contained within this null-band was taken as the simultaneous
P value of the observed partial effect curve.
© 2007 The Authors
Journal compilation © 2007 Blackwell Publishing Ltd
Cluster analysis of the total sample
The number of sampled whales at localities outside
Barrow (1–5 whales from each locality: Fig. 1) is too small
for statistically meaningful tests for geographical differentiation. Instead, we pooled all 134 individuals and tested
for potential population structure in this combined sample
using the software structure (Pritchard et al. 2000) and
baps (Corander et al. 2003). These programs implement
algorithms to cluster individuals on the basis of their
multilocus (microsatellite) genotype.
The algorithm in structure attempts to minimize deviations from Hardy–Weinberg genotype proportions within
loci and to minimize gametic phase disequilibrium among
loci, while simultaneously maximizing differences among
the specified (assumed) number of subpopulations, K
(Pritchard et al. 2000). We ran the software (version 2) with
a burn-in period of 100 000 and with 100 million Markov
chain Monte Carlo steps for different values of K, ranging
from 1 to 5. In the analysis, we assumed correlated allele
frequencies among putative subpopulations and allowed
for population mixture (the default options). Inference on
population structure was made by comparing the calculated log probability of the data given the assumed number
of subpopulations (Ln[Pr(X|K)]) for different values of K.
baps (Bayesian analysis of population structure) is similar
in scope to structure in the sense that it attempts to
uncover population structure from multilocus genotypes,
assuming Hardy–Weinberg equilibrium and gametic phase
equilibrium within the constituent subpopulations (Corander
et al. 2003). The software treats allele frequencies and number
of subpopulations as random variables and uses stochastic
optimization to infer the posterior mode of the number of
subpopulations. We used baps (version 3.2) to cluster individuals into an optimal number of subpopulations (K), and
ran the software with a predefined maximum of K = 5. The
software was supplied with a vector of 100 replicate maximum K values, and we repeated the runs 10 times in order
to check on the stability of the results. For each run, the
software reports the probabilities for different numbers of
clusters or subpopulations, K ≤ maximum K, and we averaged probabilities over the 10 runs and calculated the
standard errors of the means.
Results
The microsatellites reveal a fair amount of genetic variability
in the BCB whales (Table 1). All 10 loci are highly polymorphic
with observed number of alleles ranging from four (at
Tv16) to 11 (Tv17) in the combined sample of 134 bowheads.
The proportion of heterozygote individuals ranges from
0.4627 (at Tv16) to 0.8507 (Gata28), with an average of 0.6984.
There is little or no evidence for deviations from Hardy–
Weinberg expectations at any of these loci. Six loci display
1998 P . E . J O R D E E T A L .
Table 1 Summary statistics of genetic variability in 10 microsatellite
loci in 134 BCB bowhead whales. Alleles, observed number of
alleles at each locus; HO, proportion of heterozygote individuals;
FIS, deviations from Hardy–Weinberg genotype proportions. P
values refer to exact tests for Hardy–Weinberg disequilibrium
using genepop (Raymond & Rousset 1995), with a joint P value
calculated over loci using Fisher’s summation procedure
Locus
Alleles
HO
FIS
P value
Tv11
Tv13
Tv14
Tv16
Tv17
Tv19
Tv20
Gata28
Ev1
Ev104
Average
5
7
8
4
11
6
6
10
6
10
7.3
0.5769
0.7090
0.5821
0.4627
0.8175
0.7438
0.6791
0.8507
0.7559
0.8062
0.6984
0.0576
− 0.0209
0.0918
0.0425
−0.0184
0.0457
− 0.0265
0.0177
− 0.0047
0.0358
0.0204
0.654
0.563
0.175
0.571
0.598
0.273
0.498
0.996
0.842
0.627
0.881
positive FIS-estimates (i.e. a deficiency of heterozygotes)
and four are negative (excess heterozygotes), resulting in a
small positive average of 0.0204. Neither the average nor
any of the single-locus tests were significant, with P values
ranging from 0.175 (at Tv 14) to 0.996 (at Gata28) and 0.881
for all loci considered jointly (Table 1).
In the pairwise analysis of whales caught at Barrow
(n = 117 individuals), the average amount of genetic differentiation (equation 1) within all [n(n−1)/2 = 6786] pairs is
a = 0.0240. Of these, pairs caught during the spring season
in the same year (244 pairs) average 0.0505, and pairs
caught during the autumn season in the same year (250
pairs) average 0.0254, respectively. These average genetic
differences, being nominally larger than zero in both seasons, demonstrate that genes residing in different individuals are more often of different allelic types than are the
two genes within individuals (cf. equation 1). This reflects
the slight excess homozygosity in the sample.
Age-related effects
We find no evidence for any effect of difference in birth
year y or mean birth year m on the genetic difference
between whales, neither in the spring nor in the autumn.
This holds true both when considered separately or jointly
(Table 2: rows 2, 3, and 6). Including the age factors in
models where catch days apart, d, is present systematically
increases the P value of tests involving d.
Table 2 GAM model results for bowhead whales caught off
Barrow in the spring (244 pairs) and autumn (250 pairs), based on
10 microsatellite loci. Shown are estimated R2 = var(â)/var(a)
(variances of fitted and observed genetic difference, respectively)
and corresponding P values for R2 for different combinations of
factors aimed at explaining genetic differences within whale pairs
(equation 2): days between capture (d), years between birth (y),
and mean birth year (m). The P values are estimated from 5000
replicated randomly permuted data. For the partial temporal
effects (rows 8 and 9), only day of catch is randomly permuted
within season and year while birth year were as observed;
(NA, not applicable)
Spring
Autumn
Model
Factor
R2
P value
R2
P value
1
2
3
4
5
6
7
8
9
d
y
m
d+y
d+m
y+m
d+y+m
d|y
d|y + m
0.012
0.004
0.006
0.014
0.017
0.012
0.022
NA
NA
0.45
0.89
0.84
0.85
0.80
0.91
0.91
0.23
0.54
0.062
0.068
0.038
0.114
0.086
0.076
0.121
NA
NA
0.016
0.34
0.68
0.04
0.06
0.49
0.10
0.008
0.021
individuals on their genetic differences (Fig. 3a). The
partial effect curve rises around d = 7 (cf. Fig. 4a), similar to
that observed in the autumn (below), but the effect is weak
and nonsignificant. The simultaneous P value for the
partial effects curve in the spring is 0.392, and none of the
covariates have significant effects in this season (P values
range from 0.23 to 0.91: Table 2). During the autumn
migration, on the other hand, there is a notable effect of
days between capture (d) on genetic differences among
whales. The effect takes the form of increasing genetic
differences with days apart, from a = 0.0254 for individuals
caught on the same day to a = 0.0848 for individuals caught
5–11 days apart, and a subsequent decline in genetic
difference for individuals caught at still longer intervals
(Fig. 3b). The effect on catch days apart is apparent from
the partial effects curve (Fig. 4b), which has a simultaneous
P value of 0.009. This effect also comes up as significant
(P value = 0.016) in the GAM analysis (Table 2: row 1). The
P value rises only slightly, to 0.021, when controlling for
additive age effects, y + m (row 9). Hence, there is a significant
temporal pattern in genetic differentiation during the
autumn migration in these BCB bowhead whales.
Cluster analyses
Effects of time of capture
For whales caught during the spring migration, there is no
significant effect of days apart (d) between landings of
Disregarding any information on timing of catches and
year of birth, the analyses of potential substructure in
the pooled material (n = 134) with structure and baps
© 2007 The Authors
Journal compilation © 2007 Blackwell Publishing Ltd
B O W H E A D W H A L E G E N E T I C S T R U C T U R E 1999
Fig. 3 Genetic difference (aij, based on 10
microsatellite loci) among pairs of whales
(points) caught different number of days
apart (dij) at Barrow: (a) during spring; (b)
during autumn. The solid lines are the
estimated direct effects of days apart on
pairwise genetic differences.
yielded some, although conflicting, indications of population
substructuring in BCB bowheads (cf. Table 3). Where baps
found an optimal number of K = 4 genetically distinct
groups or populations, structure found either one or
three groups with approximately the same likelihood.
Convergence of the Markov chain algorithm in structure
was slow with these data and preliminary trial runs, based
on 1 million iterations, resulted in variable log-likelihood
values among runs for each K value (with standard
deviations of the same magnitude as the difference in the
mean log-likelihood for different K values). The results
depicted in Table 3 for this software are based on 100
million iterations, which was the largest number feasible,
but may still be unreliable.
© 2007 The Authors
Journal compilation © 2007 Blackwell Publishing Ltd
Discussion
We have detected potential population substructure
in BCB bowhead whales in the absence of adequate
geographical sample coverage. However, explaining our
finding biologically remains a challenge. Comparisons
with alternative approaches demonstrate that our method
of analysing genetic differentiation between individuals
during migration represents a viable alternative when
geographical data are unavailable or sparse, as for the
bowhead whale. The software structure obviously had
problems with slow convergence when applied to the
bowhead microsatellite data. Similar problems have been
noted in computer simulations when genetic differentiation
2000 P . E . J O R D E E T A L .
Fig. 4 Partial effects (solid line) on genetic
differences of days between landing (d) of
whales caught off Barrow: (a) during the
spring (244 pairs, simultaneous P value
0.392); and (b) during the autumn (250
pairs, simultaneous P value 0.009), when
age effects (difference in birth year, y:
Table 2 model 4) are accounted for. Simultaneous null bands of 50%, 95% and 99%
confidence are indicated with punctuated
lines and are based on 5000 replicated
permutations. (Note the differences in
scale of the vertical axes in the two figures.)
is weak (Waples & Gaggiotti 2006). Inference about K is
based on an approximation that Pritchard et al. (2000)
characterize as ‘dubious at best’ (p. 949), and they strongly
urge caution when interpreting such results. Noting this
caveat, it appears that the structure program preferred to
cluster the whales into one or three groups.
When applying baps to estimate K, we ignored the little
information that is available on geographical location of
the samples (17 of 134 individuals were caught outside
Barrow: Fig. 1). This was done because the whales were
collected during migration and little, if any, information on
population of origin is likely to be contained in the sample
locations in this case. Nevertheless, baps apparently
detected some genetic heterogeneity in the data and estimated the most likely number of subgroups to be four. It
seems biologically doubtful, however, that three or four
different populations of bowhead whales should exist in
the Bering–Chukchi–Beaufort Seas and be included in the
samples, which were largely from the single location of
Barrow. Until the performances of baps and structure are
better understood, especially with regards to potentially
false positives for baps, we choose to regard such results as
doubtful.
While our analyses with structure and baps are not
exhaustive, the results are in line with computer simulations (e.g. Waples & Gaggiotti 2006) demonstrating the
© 2007 The Authors
Journal compilation © 2007 Blackwell Publishing Ltd
B O W H E A D W H A L E G E N E T I C S T R U C T U R E 2001
Table 3 Results of cluster analyses of the total (n = 134) bowhead
sample, using structure and baps. Values in boldface indicate
the most likely number(s) of genetically different clusters of
whales, as reported by each software. structure reports the loglikelihood of each K value given the data, Ln[Pr(X|K)], whereas
baps reports the probabilities, Prob, of each K value directly
(SE, standard error, calculated over replicate runs)
No. of
clusters K
structure
Ln[Pr(X|K)]
baps
Prob (SE)
1
2
3
4
5
–3926.4
−3940.9
–3926.2
−3951.9
−3958.9
0
0
0.047 (0.017)
0.899 (0.036)
0.054 (0.030)
general difficulty, and failure of structure and baps in
particular, on reliably uncovering population structure
from genetic data alone when differentiation is not very
strong.
Our method, in contrast, utilizes auxiliary information
of potential relevance for population substructuring,
including timing of catches during migration. This approach
did indeed detect an effect-of-time of capture on the
genetic differences among individuals. The effect is manifested as an elevated pairwise genetic difference among
whales taken about a week apart (Fig. 4). This pattern is not
associated directly with any particular date in the migration but rather with the length of the interval between
when each whale in a pair passes Barrow. The effect-oftime separation in capture was significant only for whales
caught during the autumn although a similar, but not
significant, feature also appears in the spring. The time-ofcapture effect was found to remain when correcting for
birth year and age differences. Such a ‘bump’ in the distribution of genetic differentiation in whale pairs indicates a
shifting pattern of genetic similarity among individuals
during migration. If, for one of several reasons (below),
genetically similar individuals tend to comigrate, individuals caught close in time should be more similar than those
caught at longer intervals, as observed (the left part of the
bump depicted in Fig. 4b). The subsequent decline observed
at longer intervals (the right part of the bump) indicates
that groups of genetically similar individuals reappear
later, after about 2 weeks, in the migration. However, the
number of individual whale pairs falls rapidly with
increasing d (cf. Table 4), and little weight can be given to
the apparent very deep decline at large ds, as indicated by
the expanding null-bands of confidence in this region
(Fig. 4b). Possible biological explanations for the observed
bump include genetic subdivision of geographically separated breeding populations that alternate during migration; demographic phenomena, like social structuring of
the migratory population; or historical phenomena (commercial whaling) leaving a signal on the genetic composition of age classes. Both the spring and fall migrations are
known to exhibit some age segregation; the extent of social
structuring during migration is unknown.
In a single, randomly mating population, genotypes are
expected to occur in Hardy–Weinberg proportions, and
individuals should be equally similar genetically regardless of when (in the season) they were collected. Genetic
differences among samples and deviations from Hardy–
Weinberg proportions within them should be no greater
than can be attributed to sampling variation. Such expectations for an ideal population must be modified for species,
like the bowhead whale, in which age structure and overlapping generations cause deviations from strict panmixis.
Age-structured populations consist of a series of cohorts
that are born at different times and by different sets of
parents. Cohorts will therefore differ genetically to an extent
that depends on the amount of genetic drift (i.e. the effective size of the population) and the pattern of intermixing
of genes among different cohorts during reproduction
(Jorde & Ryman 1995). Both the effective size and the pattern
Table 4 Numbers of pairs and individuals in the GAM analysis of autumn whales
Days
apart
0–4
5–10
11–15
16–20
21–25
Total pairs
Total whales
Pairs in year
1990
1992
1996
1997
2000
2001
2002
0
1
0
0
0
1
2
6
0
3
1
0
10
5
29
32
15
2
0
78
13
7
5
3
0
0
15
6
5
1
0
0
0
6
4
3
0
0
0
0
3
3
46
36
17
30
7
136
17
*excluding one pair caught 42 days apart in 1995.
†excluding four unpaired individuals (two in 1995 and one each in 1998 and 1999).
© 2007 The Authors
Journal compilation © 2007 Blackwell Publishing Ltd
Total
Pairs*
Total
Whales†
96
75
38
33
7
249
50
2002 P . E . J O R D E E T A L .
of intermixing may be drastically changed during periods
of large demographic disturbances, such as seem likely to
have occurred during commercial whaling up until about
1914. Moreover, there is strong evidence that 1914 is less
than one bowhead lifetime ago.
We do not have sufficient demographic data on these
BCB bowheads for a quantitative evaluation of the effects
of age structure on temporal genetic shifts with or without
demographic disturbances. However, there is substantial
segregation by sex and age during migration (Angliss et al.
1995) and this segregation could lead to a temporal genetic
pattern during the seasons if there are genetic differences
between the sexes or between whales of different ages. We
did not find any genetic differences among sexes nor did
we find any age-related effects in our analyses, although
ageing errors could have obscured such effects. Ageing is
technically difficult in bowhead whales and the present
analyses use age estimates that are obtained indirectly,
from length information. These age estimates are subject to
considerable uncertainties. The problem is greatest for the
larger whales because there is little change in the length of
a whale once it reaches physical maturity. Reliably estimating
the age of whales that are older than age-at-physical-maturity
from a growth curve is therefore difficult. Imprecise ageing
may obscure detection of age effects on genetic differences
and could explain why we find no such effects. On the
other hand, our finding is also consistent with the notion
(Rooney et al. 1999a, 2001) that commercial whaling did
not have a persistent genetic impact on BCB bowheads.
Social structuring of the migrating population is another
possible explanation for a temporal pattern in genetic differences during migration. If related individuals migrate in
groups, genetic similarity is likely to be higher among
whale pairs that were caught near in time as compared to
those caught many days apart, as observed. Such a pattern
has also been detected in humpback whales where calves
accompany their mothers in their first year of life (Valsecchi
et al. 2002). However, the most straightforward socialstructuring hypothesis, with comigrating family or kinship
groups, is unlikely to explain the subsequent decline in
genetic differences for pairs caught more than a week apart.
More complicated temporal genetic patterns can be generated by interactions or associations between reproductive
success and environmental factors, as observed in the
humpback whale (Rosenbaum et al. 2002). Whether similar
mechanisms occur in the BCB bowhead and might generate the observed temporal pattern is unclear and remains a
subject for future studies.
Another possibility is that whales wintering in different
areas in or around the Bering Sea are genetically differentiated. If members of these putative populations time their
migration differently, or follow different migratory routes,
genetically differentiated populations may pass Barrow in
separate migratory groups or pulses. These pulses could
create a temporal genetic pattern during the migration season that is similar to the observed. Such a pulsed migration
is more effective at causing a genetic signal if the groups’
passages by Barrow are nearly nonoverlapping, if the
degree of genetic difference is large, and/or if the groups
are of approximately equal sizes. More refined hypotheses
along these lines are possible, but remain speculative until
further genetic data become available that strengthen or
weaken the notion of genetically differentiated BCB bowhead populations.
One potential practical concern in application of our
method is that the combinatorics of pairing individuals can
exaggerate sampling imbalances. For example, in our
autumn bowhead analysis, Table 4 shows the numbers of
whales and whale pairs examined in each year. Since models
are fit to points representing whale pairs, the analysis is
dominated by years 1996 and 2002 to a greater extent than
these years are actually represented in the samples. Because
different sample years are not equally represented throughout the season, an apparent temporal pattern within
seasons could result from this data patchiness if genetic
differences among individuals systematically differed
among years. Furthermore, the bowhead migration is
highly labile and depends on ice and other poorly understood factors. Until more data become available, it is difficult to preclude the possibility that our findings are driven
by something peculiar about 1996 and 2002.
In conclusion, a biological explanation for our finding
remains uncertain. Future studies should aim at attesting the hypothesis of genetically differentiated BCB
populations, by seeking to improve the geographical
coverage of bowhead samples (particularly near St
Lawrence Island and the Chukotka Peninsula), while
also maintaining focus on possible population admixture,
social structure, and better understanding of any genetic
signals caused by historical overexploitation. The temporal pattern found in this study is a critical first step in this
broad effort because our analysis has identified a genetic
signal that needs explanation.
Acknowledgements
This study was supported by the North Slope Borough (Alaska),
the US National Oceanic and Atmospheric Administration (through
the Alaska Eskimo Whaling Commission), and by the Centre
for Ecological and Evolutionary Synthesis, University of Oslo.
Collection of data for this study was only possible because of the
cooperation and support of the hunters who provided the samples
or allowed scientists to access the whales to obtain samples.
We thank J. Patton, R. Huebinger, and C. Matson for technical
assistance. Paul McNeil of the North Slope Borough GIS Division
kindly provided the BCB bowhead map (Fig. 1), produced with
support from the Barrow Arctic Science Consortium. We thank the
editor and three anonymous referees for valuable comments on an
earlier version of this study.
© 2007 The Authors
Journal compilation © 2007 Blackwell Publishing Ltd
B O W H E A D W H A L E G E N E T I C S T R U C T U R E 2003
References
Angliss RP, Hobbs RC, DeMaster DP (1995) Determination of the
sample size necessary to detect changes in length frequency distributions from a recovering population of bowhead whales.
Report of the International Whaling Commission, 45, 331–334.
Avise JC, Helfman GS, Saunders NC, Hales LS (1986)
Mitochondrial-DNA differentiation in North Atlantic eels —
population genetic consequences of an unusual life-history pattern.
Proceedings of the National Academy of Sciences, USA, 83, 4350–4354.
Beran R (1988) Balanced simultaneous confidence sets. Journal of
the American Statistical Association, 83, 679–686.
Bickham JW, Hunter DD, Matson CW, Huebinger RL, Patton JC,
George JC (2004) Genetic variability of nuclear microsatellite
loci in Bering–Chukchi–Beaufort Seas bowhead whales (Balaena
mysticetus): a test of the genetic bottleneck hypothesis. Paper SC/
56/BRG18 Presented to the Scientific Committee of the International
Whaling Commission, June 2004.
Bockstoce J (1986) Whales, Ice, and Men: the History of Whaling in the
Western Arctic. University of Washington Press, Seattle.
Bockstoce JR, Burns JJ (1993) Commercial whaling in the North
Pacific sector. In: The Bowhead Whale (eds Burns JJ, Montague JJ,
Cowles CJ), pp. 563–577. Society for Marine Mammalogy,
Lawrence, Kansas.
Bogoslovskaya LS (2003) The bowhead whale off Chukotka:
Integration of scientific and traditional knowledge. In: Indigenous Ways to the Present: Native Whaling in the Western Arctic (ed.
McCartney AP), pp. 209–254. University of Utah Press, Salt Lake
City, Utah.
Corander J, Waldmann P, Sillanpää MJ (2003) Bayesian analysis of
genetic differentiation between populations. Genetics, 163, 367–
374.
Daemen E, Cross T, Ollevier F, Volckaert FAM (2001) Analysis of
the genetic structure of European eel (Anguilla anguilla) using
microsatellite DNA and mtDNA markers. Marine Biology, 139,
755–764.
George JC, Bada J, Zeh J et al. (1999) Age and growth estimates of
bowhead whales (Balaena mysticetus) via aspartic acid racemization.
Canadian Journal of Zoology, 77, 571–580.
George JC, Zeh J, Suydam R, Clark C (2004) Abundance and
population trend (1978–2001) of western Arctic bowhead
whales surveyed near Barrow, Alaska. Marine Mammal Science,
20, 755–773.
Hoelzel AR (1994) Genetics and ecology of whales and dolphins.
Annual Review of Ecology and Systematics, 25, 377–399.
Hoelzel AR (1998) Genetic structure of cetacean populations in
sympatry, parapatry, and mixed assemblages: implications for
conservation policy. Journal of Heredity, 89, 451– 458.
Hunter DD (2005) Did bowhead whales (Balaena mysticetus) from
the Bering–Chukchi–Beaufort Seas undergo a genetic bottleneck?
A test using nuclear microsatellite loci. Master’s Thesis, Texas A
& M University, College Station, Texas.
Jorde PE, Ryman N (1995) Temporal allele frequency change and
estimation of effective size in populations with overlapping
generations. Genetics, 139, 1077–1090.
Koski WR, Davis RA, Miller GW, Withrow DE (1993) Reproduction. In: The Bowhead Whale (eds Burns JJ, Montague JJ, Cowles
CJ), pp. 239–274. Society for Marine Mammalogy, Lawrence,
Kansas.
Melnikov VV, Litovka DI, Zagrebin IA, Zelensky GM, Ainana LI
(2004) Shore-based counts of bowhead whales along the Chukotka Peninsula in May and June 1999–2001. Arctic, 57, 290–298.
© 2007 The Authors
Journal compilation © 2007 Blackwell Publishing Ltd
Moore SE, Reeves RR (1993) Distribution and movement. In:
The Bowhead Whale (eds Burns JJ, Montague JJ, Cowles CJ),
pp. 313–386. Society for Marine Mammalogy, Lawrence,
Kansas.
Nesbø CL, Rueness EK, Iversen SA, Skagen DW, Jakobsen KA
(2000) Phylogeography and population history of Atlantic
mackerel (Scomber scombrus L.): a genealogical approach
reveals genetic structuring among the eastern Atlantic
stocks. Proceedings of the Royal Society of London, Series B, Biological
Sciences, 267, 281–292.
Paetkau D, Calvert W, Stirling I, Strobeck C (1995) Microsatellite
analysis of population structure in Canadian polar bears. Molecular
Ecology, 4, 347–354.
Palsbøll P, Berube M, Larsen A, Jørgensen H (1997) Primers for the
amplification of tri- and tetramer microsatellite loci in baleen
whales. Molecular Ecology, 6, 893–895.
Pritchard JK, Stephens M, Donnelly P (2000) Inference of population structure using multilocus genotype data. Genetics, 155,
945–959.
Rannala B, Mountain JL (1997) Detecting immigration by using
multilocus genotypes. Proceedings of the National Academy of
Sciences, USA, 94, 9197–9201.
Raymond M, Rousset F (1995) genepop (version 1.2) — populationgenetics software for exact tests and ecumenicism. Journal of
Heredity, 86, 248–249.
Rooney AP, Honeycutt RL, Davis SK, Derr J (1999a) Evaluating a
putative bottleneck in a population of bowhead whales from
patterns of microsatellite diversity and genetic disequilibria.
Journal of Molecular Evolution, 49, 682–690.
Rooney AP, Honeycutt RL, Derr J (2001) Historical population size
change of bowhead whales inferred from DNA sequence
polymorphism data. Evolution, 55, 1678–1685.
Rooney AP, Merritt DB, Derr JN (1999b) Microsatellite diversity in
captive bottlenose dolphins (Tursiops truncatus). Journal of Heredity,
90, 228–231.
Rosa C, George JC, Zeh J et al. (2004) Update on Age Estimation
of Bowhead Whales (Balaena mysticetus) Using Aspartic Acid
Racemization. Paper SC/56/BRG6 presented to the Scientific Committee of the International Whaling Commission, June 2004.
Rosenbaum HC, Weinrich MT, Stoleson SA, Gibbs JP, Baker CS,
DeSalle R (2002) The effect of differential reproductive success
on population genetic structure: correlations of life history with
matrilines in humpback whales of the Gulf of Maine. Journal of
Heredity, 93, 389–399.
Rousset F (2000) Genetic differentiation between individuals.
Journal of Evolutionary Biology, 13, 58–62.
Rugh D, DeMaster D, Rooney A, Breiwick J, Shelden K, Moore S
(2003) A review of bowhead whale (Balaena mysticetus) stock
identity. Journal of Cetacean Research and Management, 5, 267–279.
Ryman N (1997) Minimizing adverse effects of fish culture: understanding the genetics of populations with overlapping generations. ICES Journal of Marine Science, 54, 1149–l159.
Smouse PE, Waples RS, Tworek JA (1990) A genetic mixture analysis for use with incomplete source population-data. Canadian
Journal of Fisheries and Aquatic Sciences, 47, 620–634.
Valsecchi E, Amos W (1996) Microsatellite markers for the study
of cetacean populations. Molecular Ecology, 5, 151–156.
Valsecchi E, Hale P, Corkeron P, Amos W (2002) Social structure
in migrating humpback whales (Megaptera novaeangliae). Molecular
Ecology, 11, 507–518.
Venables WN, Ripley BD (1999) Modern Applied Statistics with SPlus, 3rd edn. Springer, New York.
2004 P . E . J O R D E E T A L .
Waples RS, Gaggiotti O (2006) What is a population? An empirical
evaluation of some genetic methods for identifying the number
of gene pools and their degree of connectivity. Molecular Ecology,
15, 1419–1439.
Weir BS, Cockerham CC (1984) Estimating F-statistics for the analysis of population structure. Evolution, 38, 1358–1370.
Wirth T, Bernatchez L (2001) Genetic evidence against panmixia in
the European eel. Nature, 409, 1037–1040.
Zeh JE, Clark CW, George JC, Withrow D, Carroll GW, Koski WR
(1993) Current population size and dynamics. In: The Bowhead
Whale (eds Burns JJ, Montague JJ, Cowles CJ), pp. 409 – 489.
Society for Marine Mammalogy, Lawrence, Kansas.
The context of this work is the Scientific Committee of the
International Whaling Commission (IWC), where four of the
authors are members. IWC manages the aboriginal subsistence
BCB bowhead whaling. The tissue samples were collected by
North Slope Borough scientists and hunters of the Alaska Eskimo
Whaling Commission, and genotyped in Texas at the laboratory of
John Bickham. The data were made available, and the international
cooperation and collaboration thus initiated, through the Data
Availability Agreement procedure of the IWC (Journal of Cetacean
Research & Management 2004, vol. 6 (Suppl.): 406–407). Statistical
modelling and analyses were carried out in Oslo at the CEES,
headed by Nils Chr. Stenseth (http://biologi.uio.no/cees/).
© 2007 The Authors
Journal compilation © 2007 Blackwell Publishing Ltd